Work

Overjet has two lines of business. The first one caters to insurance companies and the second one to dental clinics. Both lines of business are powered by an in-house AI team of researchers. The AI is FDA approved and used to help dentists make more accurate diagnoses and patient communication. I ran the engineering team for the dental clinics line of business. My team had 10+ engineers, comprising frontend, backend, and full-stack engineers, spread across the world.

It was my first experience at a startup, and I especially enjoyed the breadth and pace of the work.

Senior Engineering Manager (February, 2022 - January, 2024)

• The dental practice software adoption grew from 30 to 2000+ clinics in two years. It was an incredible time, to work closely with early adopters, ship new features, and rebuilt the whole system for the scale.
• Hardware integration for a software team is an incredible challenge. Over time we realized that we could not bank on the quality of images taken by traditional systems at dental clinics, and we had to get closer to the source of the image, and hence had to build an integration with the scanner. We built our first hardware-integrated software to get AI diagnoses on dental x-rays as they are captured. It was launched under the name of IRIS.
• Original architecture was overly reliant on caching to run our software, but as we grew, rebuilding the cache, at very regular intervals, was not sustainable. We decided to rebuild the system to be more scalable and maintainable, leaning heavily into a event driven architecture. As patients walked in, got their x-rays taken, AI would process the images and the results would be available to the dental practitioner in near real-time.
• We had our in-house authorization and authentication system. It had to be HIPAA compliant but as we grew, and took on larger DSO customers, we realized that we needed something more robust, with features like two-factor authentication, single sign-on, etc. Classic build vs buy, we decided to go and integrate with Auth0.

I was at Chewy during a high growth phase. The company had been bought by a private equity firm and was looking to grow the business and scale rapidly. Chewy had a larger market share of the pet food market compared to Amazon, and other competitors, and the company was looking to continue to increase its market share, scale the business, as well as expand into domains like pet healthcare and pharmacy. Covid-19 pushed the adoption of online pet food even further.

My time was split in two halves. Initially, I was a senior software engineer and spent my time building an automation framework, a performance testing framework, some tools for merchandising teams to improve vendor integrations, and for the CTO office to build MVPs. Finally, in late 2018, I found my longer-term home in the rebuilding of promotions and segmentation platform, where I ended up leading a team of 8 engineers as the engineering manager.

The company was going through a rebuilding of our legacy system to break up the monolith into smaller services. Over six months, we rebuilt the promotions and segmentation platform and launched all the traffic onto the new system, in close integration with other marketing technology, storefront, checkout, order management, and customer service teams. The entire architecture was built on AWS, with active-active availability across multiple regions.

Engineering Manager (July, 2020 - February, 2022)

• Scale was always a focus and continued challenge with growth of Chewy’s customer base. Peak scale was during the 2021 holiday season, with 30 million+ customers and nearly $4 billion worth of orders going through our promotions platform without any performance issues. P95 was 50ms.
• Integration with Chewy’s pet pharmacy and pet healthcare platforms
• Rebuilt promotion’s business intelligence pipeline using Kafka

Senior Software Engineer (December, 2017 - July, 2020)

• Led the rebuilding of the promotions and segmentation platform, which was a monolith system that was broken up into smaller services using Kotlin microservices on AWS. The end goal was to have these services have a response time of 25ms, with multi-region availability, in an active-active configuration.
• 2019 Holiday season was the first time our new services were fully rolled out. It was a massive success. AWS had Kinesis failure the night before Thanksgiving in us-east-1. Even though teams in the company had to work through to get our site up and running, it was extremely satisfying to see the promotional platform run on failover to us-west-1 automatically.
• Built automation and performance testing frameworks, used across the company, as we moved from a monolith to a microservice architecture.

Cimpress was just being formed as the parent company of Vistaprint. I joined when it was a startup, as they were building out a mass customization platform. A new coffee shop might not need 100 t-shirts or large quantities of banners and other promotional items. Focus was one such small-to-medium scale business, while already catering to large-scale businesses.

The group ran as a startup but within a larger and well-established company.

Software Engineer in Test (September, 2014 - December, 2017)

• Cimpress as a platform connected to many business units, especially internationally. So for an order, a t-shirt might be routed to a different fulfiller compared to a banner or business cards. My role was to ensure that the platform could handle a variety of fulfillers and order types in a scalable and reliable way. I wrote the automation framework as the platform was being built.
• A constant pain point was that the website or the ability to place orders was not ready, but the fulfillers had fully integrated with the platform. I built an internal tool to place orders of all types, choosing the fulfiller without needing the website. It was an exciting project, which helped parallelize the launch of many new products and merchants.
• Built a performance testing framework using Locust for Vistaprint to prepare for the holiday season. This created synthetic load, mimicking real traffic, and allowed the teams to test the system under load.

I was part of Pearson Education’s Identity and Access Management team, building Rumba.

Automation Engineer (January, 2012 - September, 2014)

• Led end-to-end test automation initiatives for Identity and Access Management platform migration, dramatically improving test coverage from 20% to 80% through development of comprehensive SOAP/REST API and Selenium-based UI test frameworks
• Established quality assurance processes including test planning, acceptance criteria definition, and cross-platform integration testing while serving as key liaison between engineering teams and stakeholders

Right after my masters, I joined Geode Capital Management, LLC as a research intern. It was a small group and they had nearly $1.5 billion in assets under management at the time.

Research Intern (August, 2010 - October, 2010)

• Built and worked on research projects and strategies for the firm, using R for statistical analysis
• Most exciting for me at that time was to go through periodic investment journals and find new ideas and strategies, research them and present to fellow research team and portfolio managers. Of all the strategies presented, my work on pairs trading was the most successful